1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/filter.h> 60 #include <linux/rculist_nulls.h> 61 #include <linux/poll.h> 62 #include <linux/sockptr.h> 63 #include <linux/indirect_call_wrapper.h> 64 #include <linux/atomic.h> 65 #include <linux/refcount.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 165 * address on 64bit arches : cf INET_MATCH() 166 */ 167 union { 168 __addrpair skc_addrpair; 169 struct { 170 __be32 skc_daddr; 171 __be32 skc_rcv_saddr; 172 }; 173 }; 174 union { 175 unsigned int skc_hash; 176 __u16 skc_u16hashes[2]; 177 }; 178 /* skc_dport && skc_num must be grouped as well */ 179 union { 180 __portpair skc_portpair; 181 struct { 182 __be16 skc_dport; 183 __u16 skc_num; 184 }; 185 }; 186 187 unsigned short skc_family; 188 volatile unsigned char skc_state; 189 unsigned char skc_reuse:4; 190 unsigned char skc_reuseport:1; 191 unsigned char skc_ipv6only:1; 192 unsigned char skc_net_refcnt:1; 193 int skc_bound_dev_if; 194 union { 195 struct hlist_node skc_bind_node; 196 struct hlist_node skc_portaddr_node; 197 }; 198 struct proto *skc_prot; 199 possible_net_t skc_net; 200 201 #if IS_ENABLED(CONFIG_IPV6) 202 struct in6_addr skc_v6_daddr; 203 struct in6_addr skc_v6_rcv_saddr; 204 #endif 205 206 atomic64_t skc_cookie; 207 208 /* following fields are padding to force 209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 210 * assuming IPV6 is enabled. We use this padding differently 211 * for different kind of 'sockets' 212 */ 213 union { 214 unsigned long skc_flags; 215 struct sock *skc_listener; /* request_sock */ 216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 217 }; 218 /* 219 * fields between dontcopy_begin/dontcopy_end 220 * are not copied in sock_copy() 221 */ 222 /* private: */ 223 int skc_dontcopy_begin[0]; 224 /* public: */ 225 union { 226 struct hlist_node skc_node; 227 struct hlist_nulls_node skc_nulls_node; 228 }; 229 unsigned short skc_tx_queue_mapping; 230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 231 unsigned short skc_rx_queue_mapping; 232 #endif 233 union { 234 int skc_incoming_cpu; 235 u32 skc_rcv_wnd; 236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 237 }; 238 239 refcount_t skc_refcnt; 240 /* private: */ 241 int skc_dontcopy_end[0]; 242 union { 243 u32 skc_rxhash; 244 u32 skc_window_clamp; 245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 246 }; 247 /* public: */ 248 }; 249 250 struct bpf_local_storage; 251 252 /** 253 * struct sock - network layer representation of sockets 254 * @__sk_common: shared layout with inet_timewait_sock 255 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 256 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 257 * @sk_lock: synchronizer 258 * @sk_kern_sock: True if sock is using kernel lock classes 259 * @sk_rcvbuf: size of receive buffer in bytes 260 * @sk_wq: sock wait queue and async head 261 * @sk_rx_dst: receive input route used by early demux 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_rx_skb_cache: cache copy of recently accessed RX skb 266 * @sk_receive_queue: incoming packets 267 * @sk_wmem_alloc: transmit queue bytes committed 268 * @sk_tsq_flags: TCP Small Queues flags 269 * @sk_write_queue: Packet sending queue 270 * @sk_omem_alloc: "o" is "option" or "other" 271 * @sk_wmem_queued: persistent queue size 272 * @sk_forward_alloc: space allocated forward 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 286 * @sk_route_forced_caps: static, forced route capabilities 287 * (set in tcp_init_sock()) 288 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 289 * @sk_gso_max_size: Maximum GSO segment size to build 290 * @sk_gso_max_segs: Maximum number of GSO segments 291 * @sk_pacing_shift: scaling factor for TCP Small Queues 292 * @sk_lingertime: %SO_LINGER l_linger setting 293 * @sk_backlog: always used with the per-socket spinlock held 294 * @sk_callback_lock: used with the callbacks in the end of this struct 295 * @sk_error_queue: rarely used 296 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 297 * IPV6_ADDRFORM for instance) 298 * @sk_err: last error 299 * @sk_err_soft: errors that don't cause failure but are the cause of a 300 * persistent failure not just 'timed out' 301 * @sk_drops: raw/udp drops counter 302 * @sk_ack_backlog: current listen backlog 303 * @sk_max_ack_backlog: listen backlog set in listen() 304 * @sk_uid: user id of owner 305 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 306 * @sk_busy_poll_budget: napi processing budget when busypolling 307 * @sk_priority: %SO_PRIORITY setting 308 * @sk_type: socket type (%SOCK_STREAM, etc) 309 * @sk_protocol: which protocol this socket belongs in this network family 310 * @sk_peer_pid: &struct pid for this socket's peer 311 * @sk_peer_cred: %SO_PEERCRED setting 312 * @sk_rcvlowat: %SO_RCVLOWAT setting 313 * @sk_rcvtimeo: %SO_RCVTIMEO setting 314 * @sk_sndtimeo: %SO_SNDTIMEO setting 315 * @sk_txhash: computed flow hash for use on transmit 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 322 * for timestamping 323 * @sk_tskey: counter to disambiguate concurrent tstamp requests 324 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 325 * @sk_socket: Identd and reporting IO signals 326 * @sk_user_data: RPC layer private data 327 * @sk_frag: cached page frag 328 * @sk_peek_off: current peek_offset value 329 * @sk_send_head: front of stuff to transmit 330 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 331 * @sk_tx_skb_cache: cache copy of recently accessed TX skb 332 * @sk_security: used by security modules 333 * @sk_mark: generic packet mark 334 * @sk_cgrp_data: cgroup data for this cgroup 335 * @sk_memcg: this socket's memory cgroup association 336 * @sk_write_pending: a write to stream socket waits to start 337 * @sk_state_change: callback to indicate change in the state of the sock 338 * @sk_data_ready: callback to indicate there is data to be processed 339 * @sk_write_space: callback to indicate there is bf sending space available 340 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 341 * @sk_backlog_rcv: callback to process the backlog 342 * @sk_validate_xmit_skb: ptr to an optional validate function 343 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 344 * @sk_reuseport_cb: reuseport group container 345 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 346 * @sk_rcu: used during RCU grace period 347 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 348 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 349 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 350 * @sk_txtime_unused: unused txtime flags 351 */ 352 struct sock { 353 /* 354 * Now struct inet_timewait_sock also uses sock_common, so please just 355 * don't add nothing before this first member (__sk_common) --acme 356 */ 357 struct sock_common __sk_common; 358 #define sk_node __sk_common.skc_node 359 #define sk_nulls_node __sk_common.skc_nulls_node 360 #define sk_refcnt __sk_common.skc_refcnt 361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 364 #endif 365 366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 368 #define sk_hash __sk_common.skc_hash 369 #define sk_portpair __sk_common.skc_portpair 370 #define sk_num __sk_common.skc_num 371 #define sk_dport __sk_common.skc_dport 372 #define sk_addrpair __sk_common.skc_addrpair 373 #define sk_daddr __sk_common.skc_daddr 374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 375 #define sk_family __sk_common.skc_family 376 #define sk_state __sk_common.skc_state 377 #define sk_reuse __sk_common.skc_reuse 378 #define sk_reuseport __sk_common.skc_reuseport 379 #define sk_ipv6only __sk_common.skc_ipv6only 380 #define sk_net_refcnt __sk_common.skc_net_refcnt 381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 382 #define sk_bind_node __sk_common.skc_bind_node 383 #define sk_prot __sk_common.skc_prot 384 #define sk_net __sk_common.skc_net 385 #define sk_v6_daddr __sk_common.skc_v6_daddr 386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 387 #define sk_cookie __sk_common.skc_cookie 388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 389 #define sk_flags __sk_common.skc_flags 390 #define sk_rxhash __sk_common.skc_rxhash 391 392 socket_lock_t sk_lock; 393 atomic_t sk_drops; 394 int sk_rcvlowat; 395 struct sk_buff_head sk_error_queue; 396 struct sk_buff *sk_rx_skb_cache; 397 struct sk_buff_head sk_receive_queue; 398 /* 399 * The backlog queue is special, it is always used with 400 * the per-socket spinlock held and requires low latency 401 * access. Therefore we special case it's implementation. 402 * Note : rmem_alloc is in this structure to fill a hole 403 * on 64bit arches, not because its logically part of 404 * backlog. 405 */ 406 struct { 407 atomic_t rmem_alloc; 408 int len; 409 struct sk_buff *head; 410 struct sk_buff *tail; 411 } sk_backlog; 412 #define sk_rmem_alloc sk_backlog.rmem_alloc 413 414 int sk_forward_alloc; 415 #ifdef CONFIG_NET_RX_BUSY_POLL 416 unsigned int sk_ll_usec; 417 /* ===== mostly read cache line ===== */ 418 unsigned int sk_napi_id; 419 #endif 420 int sk_rcvbuf; 421 422 struct sk_filter __rcu *sk_filter; 423 union { 424 struct socket_wq __rcu *sk_wq; 425 /* private: */ 426 struct socket_wq *sk_wq_raw; 427 /* public: */ 428 }; 429 #ifdef CONFIG_XFRM 430 struct xfrm_policy __rcu *sk_policy[2]; 431 #endif 432 struct dst_entry *sk_rx_dst; 433 struct dst_entry __rcu *sk_dst_cache; 434 atomic_t sk_omem_alloc; 435 int sk_sndbuf; 436 437 /* ===== cache line for TX ===== */ 438 int sk_wmem_queued; 439 refcount_t sk_wmem_alloc; 440 unsigned long sk_tsq_flags; 441 union { 442 struct sk_buff *sk_send_head; 443 struct rb_root tcp_rtx_queue; 444 }; 445 struct sk_buff *sk_tx_skb_cache; 446 struct sk_buff_head sk_write_queue; 447 __s32 sk_peek_off; 448 int sk_write_pending; 449 __u32 sk_dst_pending_confirm; 450 u32 sk_pacing_status; /* see enum sk_pacing */ 451 long sk_sndtimeo; 452 struct timer_list sk_timer; 453 __u32 sk_priority; 454 __u32 sk_mark; 455 unsigned long sk_pacing_rate; /* bytes per second */ 456 unsigned long sk_max_pacing_rate; 457 struct page_frag sk_frag; 458 netdev_features_t sk_route_caps; 459 netdev_features_t sk_route_nocaps; 460 netdev_features_t sk_route_forced_caps; 461 int sk_gso_type; 462 unsigned int sk_gso_max_size; 463 gfp_t sk_allocation; 464 __u32 sk_txhash; 465 466 /* 467 * Because of non atomicity rules, all 468 * changes are protected by socket lock. 469 */ 470 u8 sk_padding : 1, 471 sk_kern_sock : 1, 472 sk_no_check_tx : 1, 473 sk_no_check_rx : 1, 474 sk_userlocks : 4; 475 u8 sk_pacing_shift; 476 u16 sk_type; 477 u16 sk_protocol; 478 u16 sk_gso_max_segs; 479 unsigned long sk_lingertime; 480 struct proto *sk_prot_creator; 481 rwlock_t sk_callback_lock; 482 int sk_err, 483 sk_err_soft; 484 u32 sk_ack_backlog; 485 u32 sk_max_ack_backlog; 486 kuid_t sk_uid; 487 #ifdef CONFIG_NET_RX_BUSY_POLL 488 u8 sk_prefer_busy_poll; 489 u16 sk_busy_poll_budget; 490 #endif 491 struct pid *sk_peer_pid; 492 const struct cred *sk_peer_cred; 493 long sk_rcvtimeo; 494 ktime_t sk_stamp; 495 #if BITS_PER_LONG==32 496 seqlock_t sk_stamp_seq; 497 #endif 498 u16 sk_tsflags; 499 int sk_bind_phc; 500 u8 sk_shutdown; 501 u32 sk_tskey; 502 atomic_t sk_zckey; 503 504 u8 sk_clockid; 505 u8 sk_txtime_deadline_mode : 1, 506 sk_txtime_report_errors : 1, 507 sk_txtime_unused : 6; 508 509 struct socket *sk_socket; 510 void *sk_user_data; 511 #ifdef CONFIG_SECURITY 512 void *sk_security; 513 #endif 514 struct sock_cgroup_data sk_cgrp_data; 515 struct mem_cgroup *sk_memcg; 516 void (*sk_state_change)(struct sock *sk); 517 void (*sk_data_ready)(struct sock *sk); 518 void (*sk_write_space)(struct sock *sk); 519 void (*sk_error_report)(struct sock *sk); 520 int (*sk_backlog_rcv)(struct sock *sk, 521 struct sk_buff *skb); 522 #ifdef CONFIG_SOCK_VALIDATE_XMIT 523 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 524 struct net_device *dev, 525 struct sk_buff *skb); 526 #endif 527 void (*sk_destruct)(struct sock *sk); 528 struct sock_reuseport __rcu *sk_reuseport_cb; 529 #ifdef CONFIG_BPF_SYSCALL 530 struct bpf_local_storage __rcu *sk_bpf_storage; 531 #endif 532 struct rcu_head sk_rcu; 533 }; 534 535 enum sk_pacing { 536 SK_PACING_NONE = 0, 537 SK_PACING_NEEDED = 1, 538 SK_PACING_FQ = 2, 539 }; 540 541 /* Pointer stored in sk_user_data might not be suitable for copying 542 * when cloning the socket. For instance, it can point to a reference 543 * counted object. sk_user_data bottom bit is set if pointer must not 544 * be copied. 545 */ 546 #define SK_USER_DATA_NOCOPY 1UL 547 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 548 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 549 550 /** 551 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 552 * @sk: socket 553 */ 554 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 555 { 556 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 557 } 558 559 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 560 561 #define rcu_dereference_sk_user_data(sk) \ 562 ({ \ 563 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 564 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 565 }) 566 #define rcu_assign_sk_user_data(sk, ptr) \ 567 ({ \ 568 uintptr_t __tmp = (uintptr_t)(ptr); \ 569 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 570 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 571 }) 572 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 573 ({ \ 574 uintptr_t __tmp = (uintptr_t)(ptr); \ 575 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 576 rcu_assign_pointer(__sk_user_data((sk)), \ 577 __tmp | SK_USER_DATA_NOCOPY); \ 578 }) 579 580 /* 581 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 582 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 583 * on a socket means that the socket will reuse everybody else's port 584 * without looking at the other's sk_reuse value. 585 */ 586 587 #define SK_NO_REUSE 0 588 #define SK_CAN_REUSE 1 589 #define SK_FORCE_REUSE 2 590 591 int sk_set_peek_off(struct sock *sk, int val); 592 593 static inline int sk_peek_offset(struct sock *sk, int flags) 594 { 595 if (unlikely(flags & MSG_PEEK)) { 596 return READ_ONCE(sk->sk_peek_off); 597 } 598 599 return 0; 600 } 601 602 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 603 { 604 s32 off = READ_ONCE(sk->sk_peek_off); 605 606 if (unlikely(off >= 0)) { 607 off = max_t(s32, off - val, 0); 608 WRITE_ONCE(sk->sk_peek_off, off); 609 } 610 } 611 612 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 613 { 614 sk_peek_offset_bwd(sk, -val); 615 } 616 617 /* 618 * Hashed lists helper routines 619 */ 620 static inline struct sock *sk_entry(const struct hlist_node *node) 621 { 622 return hlist_entry(node, struct sock, sk_node); 623 } 624 625 static inline struct sock *__sk_head(const struct hlist_head *head) 626 { 627 return hlist_entry(head->first, struct sock, sk_node); 628 } 629 630 static inline struct sock *sk_head(const struct hlist_head *head) 631 { 632 return hlist_empty(head) ? NULL : __sk_head(head); 633 } 634 635 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 636 { 637 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 638 } 639 640 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 641 { 642 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 643 } 644 645 static inline struct sock *sk_next(const struct sock *sk) 646 { 647 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 648 } 649 650 static inline struct sock *sk_nulls_next(const struct sock *sk) 651 { 652 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 653 hlist_nulls_entry(sk->sk_nulls_node.next, 654 struct sock, sk_nulls_node) : 655 NULL; 656 } 657 658 static inline bool sk_unhashed(const struct sock *sk) 659 { 660 return hlist_unhashed(&sk->sk_node); 661 } 662 663 static inline bool sk_hashed(const struct sock *sk) 664 { 665 return !sk_unhashed(sk); 666 } 667 668 static inline void sk_node_init(struct hlist_node *node) 669 { 670 node->pprev = NULL; 671 } 672 673 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 674 { 675 node->pprev = NULL; 676 } 677 678 static inline void __sk_del_node(struct sock *sk) 679 { 680 __hlist_del(&sk->sk_node); 681 } 682 683 /* NB: equivalent to hlist_del_init_rcu */ 684 static inline bool __sk_del_node_init(struct sock *sk) 685 { 686 if (sk_hashed(sk)) { 687 __sk_del_node(sk); 688 sk_node_init(&sk->sk_node); 689 return true; 690 } 691 return false; 692 } 693 694 /* Grab socket reference count. This operation is valid only 695 when sk is ALREADY grabbed f.e. it is found in hash table 696 or a list and the lookup is made under lock preventing hash table 697 modifications. 698 */ 699 700 static __always_inline void sock_hold(struct sock *sk) 701 { 702 refcount_inc(&sk->sk_refcnt); 703 } 704 705 /* Ungrab socket in the context, which assumes that socket refcnt 706 cannot hit zero, f.e. it is true in context of any socketcall. 707 */ 708 static __always_inline void __sock_put(struct sock *sk) 709 { 710 refcount_dec(&sk->sk_refcnt); 711 } 712 713 static inline bool sk_del_node_init(struct sock *sk) 714 { 715 bool rc = __sk_del_node_init(sk); 716 717 if (rc) { 718 /* paranoid for a while -acme */ 719 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 720 __sock_put(sk); 721 } 722 return rc; 723 } 724 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 725 726 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 727 { 728 if (sk_hashed(sk)) { 729 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 730 return true; 731 } 732 return false; 733 } 734 735 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 736 { 737 bool rc = __sk_nulls_del_node_init_rcu(sk); 738 739 if (rc) { 740 /* paranoid for a while -acme */ 741 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 742 __sock_put(sk); 743 } 744 return rc; 745 } 746 747 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 748 { 749 hlist_add_head(&sk->sk_node, list); 750 } 751 752 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 753 { 754 sock_hold(sk); 755 __sk_add_node(sk, list); 756 } 757 758 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 759 { 760 sock_hold(sk); 761 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 762 sk->sk_family == AF_INET6) 763 hlist_add_tail_rcu(&sk->sk_node, list); 764 else 765 hlist_add_head_rcu(&sk->sk_node, list); 766 } 767 768 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 769 { 770 sock_hold(sk); 771 hlist_add_tail_rcu(&sk->sk_node, list); 772 } 773 774 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 775 { 776 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 777 } 778 779 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 780 { 781 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 782 } 783 784 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 785 { 786 sock_hold(sk); 787 __sk_nulls_add_node_rcu(sk, list); 788 } 789 790 static inline void __sk_del_bind_node(struct sock *sk) 791 { 792 __hlist_del(&sk->sk_bind_node); 793 } 794 795 static inline void sk_add_bind_node(struct sock *sk, 796 struct hlist_head *list) 797 { 798 hlist_add_head(&sk->sk_bind_node, list); 799 } 800 801 #define sk_for_each(__sk, list) \ 802 hlist_for_each_entry(__sk, list, sk_node) 803 #define sk_for_each_rcu(__sk, list) \ 804 hlist_for_each_entry_rcu(__sk, list, sk_node) 805 #define sk_nulls_for_each(__sk, node, list) \ 806 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 807 #define sk_nulls_for_each_rcu(__sk, node, list) \ 808 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 809 #define sk_for_each_from(__sk) \ 810 hlist_for_each_entry_from(__sk, sk_node) 811 #define sk_nulls_for_each_from(__sk, node) \ 812 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 813 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 814 #define sk_for_each_safe(__sk, tmp, list) \ 815 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 816 #define sk_for_each_bound(__sk, list) \ 817 hlist_for_each_entry(__sk, list, sk_bind_node) 818 819 /** 820 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 821 * @tpos: the type * to use as a loop cursor. 822 * @pos: the &struct hlist_node to use as a loop cursor. 823 * @head: the head for your list. 824 * @offset: offset of hlist_node within the struct. 825 * 826 */ 827 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 828 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 829 pos != NULL && \ 830 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 831 pos = rcu_dereference(hlist_next_rcu(pos))) 832 833 static inline struct user_namespace *sk_user_ns(struct sock *sk) 834 { 835 /* Careful only use this in a context where these parameters 836 * can not change and must all be valid, such as recvmsg from 837 * userspace. 838 */ 839 return sk->sk_socket->file->f_cred->user_ns; 840 } 841 842 /* Sock flags */ 843 enum sock_flags { 844 SOCK_DEAD, 845 SOCK_DONE, 846 SOCK_URGINLINE, 847 SOCK_KEEPOPEN, 848 SOCK_LINGER, 849 SOCK_DESTROY, 850 SOCK_BROADCAST, 851 SOCK_TIMESTAMP, 852 SOCK_ZAPPED, 853 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 854 SOCK_DBG, /* %SO_DEBUG setting */ 855 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 856 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 857 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 858 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 859 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 860 SOCK_FASYNC, /* fasync() active */ 861 SOCK_RXQ_OVFL, 862 SOCK_ZEROCOPY, /* buffers from userspace */ 863 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 864 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 865 * Will use last 4 bytes of packet sent from 866 * user-space instead. 867 */ 868 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 869 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 870 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 871 SOCK_TXTIME, 872 SOCK_XDP, /* XDP is attached */ 873 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 874 }; 875 876 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 877 878 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 879 { 880 nsk->sk_flags = osk->sk_flags; 881 } 882 883 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 884 { 885 __set_bit(flag, &sk->sk_flags); 886 } 887 888 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 889 { 890 __clear_bit(flag, &sk->sk_flags); 891 } 892 893 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 894 int valbool) 895 { 896 if (valbool) 897 sock_set_flag(sk, bit); 898 else 899 sock_reset_flag(sk, bit); 900 } 901 902 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 903 { 904 return test_bit(flag, &sk->sk_flags); 905 } 906 907 #ifdef CONFIG_NET 908 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 909 static inline int sk_memalloc_socks(void) 910 { 911 return static_branch_unlikely(&memalloc_socks_key); 912 } 913 914 void __receive_sock(struct file *file); 915 #else 916 917 static inline int sk_memalloc_socks(void) 918 { 919 return 0; 920 } 921 922 static inline void __receive_sock(struct file *file) 923 { } 924 #endif 925 926 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 927 { 928 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 929 } 930 931 static inline void sk_acceptq_removed(struct sock *sk) 932 { 933 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 934 } 935 936 static inline void sk_acceptq_added(struct sock *sk) 937 { 938 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 939 } 940 941 /* Note: If you think the test should be: 942 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 943 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 944 */ 945 static inline bool sk_acceptq_is_full(const struct sock *sk) 946 { 947 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 948 } 949 950 /* 951 * Compute minimal free write space needed to queue new packets. 952 */ 953 static inline int sk_stream_min_wspace(const struct sock *sk) 954 { 955 return READ_ONCE(sk->sk_wmem_queued) >> 1; 956 } 957 958 static inline int sk_stream_wspace(const struct sock *sk) 959 { 960 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 961 } 962 963 static inline void sk_wmem_queued_add(struct sock *sk, int val) 964 { 965 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 966 } 967 968 void sk_stream_write_space(struct sock *sk); 969 970 /* OOB backlog add */ 971 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 972 { 973 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 974 skb_dst_force(skb); 975 976 if (!sk->sk_backlog.tail) 977 WRITE_ONCE(sk->sk_backlog.head, skb); 978 else 979 sk->sk_backlog.tail->next = skb; 980 981 WRITE_ONCE(sk->sk_backlog.tail, skb); 982 skb->next = NULL; 983 } 984 985 /* 986 * Take into account size of receive queue and backlog queue 987 * Do not take into account this skb truesize, 988 * to allow even a single big packet to come. 989 */ 990 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 991 { 992 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 993 994 return qsize > limit; 995 } 996 997 /* The per-socket spinlock must be held here. */ 998 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 999 unsigned int limit) 1000 { 1001 if (sk_rcvqueues_full(sk, limit)) 1002 return -ENOBUFS; 1003 1004 /* 1005 * If the skb was allocated from pfmemalloc reserves, only 1006 * allow SOCK_MEMALLOC sockets to use it as this socket is 1007 * helping free memory 1008 */ 1009 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1010 return -ENOMEM; 1011 1012 __sk_add_backlog(sk, skb); 1013 sk->sk_backlog.len += skb->truesize; 1014 return 0; 1015 } 1016 1017 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1018 1019 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1020 { 1021 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1022 return __sk_backlog_rcv(sk, skb); 1023 1024 return sk->sk_backlog_rcv(sk, skb); 1025 } 1026 1027 static inline void sk_incoming_cpu_update(struct sock *sk) 1028 { 1029 int cpu = raw_smp_processor_id(); 1030 1031 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1032 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1033 } 1034 1035 static inline void sock_rps_record_flow_hash(__u32 hash) 1036 { 1037 #ifdef CONFIG_RPS 1038 struct rps_sock_flow_table *sock_flow_table; 1039 1040 rcu_read_lock(); 1041 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1042 rps_record_sock_flow(sock_flow_table, hash); 1043 rcu_read_unlock(); 1044 #endif 1045 } 1046 1047 static inline void sock_rps_record_flow(const struct sock *sk) 1048 { 1049 #ifdef CONFIG_RPS 1050 if (static_branch_unlikely(&rfs_needed)) { 1051 /* Reading sk->sk_rxhash might incur an expensive cache line 1052 * miss. 1053 * 1054 * TCP_ESTABLISHED does cover almost all states where RFS 1055 * might be useful, and is cheaper [1] than testing : 1056 * IPv4: inet_sk(sk)->inet_daddr 1057 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1058 * OR an additional socket flag 1059 * [1] : sk_state and sk_prot are in the same cache line. 1060 */ 1061 if (sk->sk_state == TCP_ESTABLISHED) 1062 sock_rps_record_flow_hash(sk->sk_rxhash); 1063 } 1064 #endif 1065 } 1066 1067 static inline void sock_rps_save_rxhash(struct sock *sk, 1068 const struct sk_buff *skb) 1069 { 1070 #ifdef CONFIG_RPS 1071 if (unlikely(sk->sk_rxhash != skb->hash)) 1072 sk->sk_rxhash = skb->hash; 1073 #endif 1074 } 1075 1076 static inline void sock_rps_reset_rxhash(struct sock *sk) 1077 { 1078 #ifdef CONFIG_RPS 1079 sk->sk_rxhash = 0; 1080 #endif 1081 } 1082 1083 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1084 ({ int __rc; \ 1085 release_sock(__sk); \ 1086 __rc = __condition; \ 1087 if (!__rc) { \ 1088 *(__timeo) = wait_woken(__wait, \ 1089 TASK_INTERRUPTIBLE, \ 1090 *(__timeo)); \ 1091 } \ 1092 sched_annotate_sleep(); \ 1093 lock_sock(__sk); \ 1094 __rc = __condition; \ 1095 __rc; \ 1096 }) 1097 1098 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1099 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1100 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1101 int sk_stream_error(struct sock *sk, int flags, int err); 1102 void sk_stream_kill_queues(struct sock *sk); 1103 void sk_set_memalloc(struct sock *sk); 1104 void sk_clear_memalloc(struct sock *sk); 1105 1106 void __sk_flush_backlog(struct sock *sk); 1107 1108 static inline bool sk_flush_backlog(struct sock *sk) 1109 { 1110 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1111 __sk_flush_backlog(sk); 1112 return true; 1113 } 1114 return false; 1115 } 1116 1117 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1118 1119 struct request_sock_ops; 1120 struct timewait_sock_ops; 1121 struct inet_hashinfo; 1122 struct raw_hashinfo; 1123 struct smc_hashinfo; 1124 struct module; 1125 struct sk_psock; 1126 1127 /* 1128 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1129 * un-modified. Special care is taken when initializing object to zero. 1130 */ 1131 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1132 { 1133 if (offsetof(struct sock, sk_node.next) != 0) 1134 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1135 memset(&sk->sk_node.pprev, 0, 1136 size - offsetof(struct sock, sk_node.pprev)); 1137 } 1138 1139 /* Networking protocol blocks we attach to sockets. 1140 * socket layer -> transport layer interface 1141 */ 1142 struct proto { 1143 void (*close)(struct sock *sk, 1144 long timeout); 1145 int (*pre_connect)(struct sock *sk, 1146 struct sockaddr *uaddr, 1147 int addr_len); 1148 int (*connect)(struct sock *sk, 1149 struct sockaddr *uaddr, 1150 int addr_len); 1151 int (*disconnect)(struct sock *sk, int flags); 1152 1153 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1154 bool kern); 1155 1156 int (*ioctl)(struct sock *sk, int cmd, 1157 unsigned long arg); 1158 int (*init)(struct sock *sk); 1159 void (*destroy)(struct sock *sk); 1160 void (*shutdown)(struct sock *sk, int how); 1161 int (*setsockopt)(struct sock *sk, int level, 1162 int optname, sockptr_t optval, 1163 unsigned int optlen); 1164 int (*getsockopt)(struct sock *sk, int level, 1165 int optname, char __user *optval, 1166 int __user *option); 1167 void (*keepalive)(struct sock *sk, int valbool); 1168 #ifdef CONFIG_COMPAT 1169 int (*compat_ioctl)(struct sock *sk, 1170 unsigned int cmd, unsigned long arg); 1171 #endif 1172 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1173 size_t len); 1174 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1175 size_t len, int noblock, int flags, 1176 int *addr_len); 1177 int (*sendpage)(struct sock *sk, struct page *page, 1178 int offset, size_t size, int flags); 1179 int (*bind)(struct sock *sk, 1180 struct sockaddr *addr, int addr_len); 1181 int (*bind_add)(struct sock *sk, 1182 struct sockaddr *addr, int addr_len); 1183 1184 int (*backlog_rcv) (struct sock *sk, 1185 struct sk_buff *skb); 1186 bool (*bpf_bypass_getsockopt)(int level, 1187 int optname); 1188 1189 void (*release_cb)(struct sock *sk); 1190 1191 /* Keeping track of sk's, looking them up, and port selection methods. */ 1192 int (*hash)(struct sock *sk); 1193 void (*unhash)(struct sock *sk); 1194 void (*rehash)(struct sock *sk); 1195 int (*get_port)(struct sock *sk, unsigned short snum); 1196 #ifdef CONFIG_BPF_SYSCALL 1197 int (*psock_update_sk_prot)(struct sock *sk, 1198 struct sk_psock *psock, 1199 bool restore); 1200 #endif 1201 1202 /* Keeping track of sockets in use */ 1203 #ifdef CONFIG_PROC_FS 1204 unsigned int inuse_idx; 1205 #endif 1206 1207 bool (*stream_memory_free)(const struct sock *sk, int wake); 1208 bool (*stream_memory_read)(const struct sock *sk); 1209 /* Memory pressure */ 1210 void (*enter_memory_pressure)(struct sock *sk); 1211 void (*leave_memory_pressure)(struct sock *sk); 1212 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1213 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1214 /* 1215 * Pressure flag: try to collapse. 1216 * Technical note: it is used by multiple contexts non atomically. 1217 * All the __sk_mem_schedule() is of this nature: accounting 1218 * is strict, actions are advisory and have some latency. 1219 */ 1220 unsigned long *memory_pressure; 1221 long *sysctl_mem; 1222 1223 int *sysctl_wmem; 1224 int *sysctl_rmem; 1225 u32 sysctl_wmem_offset; 1226 u32 sysctl_rmem_offset; 1227 1228 int max_header; 1229 bool no_autobind; 1230 1231 struct kmem_cache *slab; 1232 unsigned int obj_size; 1233 slab_flags_t slab_flags; 1234 unsigned int useroffset; /* Usercopy region offset */ 1235 unsigned int usersize; /* Usercopy region size */ 1236 1237 struct percpu_counter *orphan_count; 1238 1239 struct request_sock_ops *rsk_prot; 1240 struct timewait_sock_ops *twsk_prot; 1241 1242 union { 1243 struct inet_hashinfo *hashinfo; 1244 struct udp_table *udp_table; 1245 struct raw_hashinfo *raw_hash; 1246 struct smc_hashinfo *smc_hash; 1247 } h; 1248 1249 struct module *owner; 1250 1251 char name[32]; 1252 1253 struct list_head node; 1254 #ifdef SOCK_REFCNT_DEBUG 1255 atomic_t socks; 1256 #endif 1257 int (*diag_destroy)(struct sock *sk, int err); 1258 } __randomize_layout; 1259 1260 int proto_register(struct proto *prot, int alloc_slab); 1261 void proto_unregister(struct proto *prot); 1262 int sock_load_diag_module(int family, int protocol); 1263 1264 #ifdef SOCK_REFCNT_DEBUG 1265 static inline void sk_refcnt_debug_inc(struct sock *sk) 1266 { 1267 atomic_inc(&sk->sk_prot->socks); 1268 } 1269 1270 static inline void sk_refcnt_debug_dec(struct sock *sk) 1271 { 1272 atomic_dec(&sk->sk_prot->socks); 1273 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1274 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1275 } 1276 1277 static inline void sk_refcnt_debug_release(const struct sock *sk) 1278 { 1279 if (refcount_read(&sk->sk_refcnt) != 1) 1280 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1281 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1282 } 1283 #else /* SOCK_REFCNT_DEBUG */ 1284 #define sk_refcnt_debug_inc(sk) do { } while (0) 1285 #define sk_refcnt_debug_dec(sk) do { } while (0) 1286 #define sk_refcnt_debug_release(sk) do { } while (0) 1287 #endif /* SOCK_REFCNT_DEBUG */ 1288 1289 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1290 1291 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1292 { 1293 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1294 return false; 1295 1296 #ifdef CONFIG_INET 1297 return sk->sk_prot->stream_memory_free ? 1298 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free, 1299 tcp_stream_memory_free, 1300 sk, wake) : true; 1301 #else 1302 return sk->sk_prot->stream_memory_free ? 1303 sk->sk_prot->stream_memory_free(sk, wake) : true; 1304 #endif 1305 } 1306 1307 static inline bool sk_stream_memory_free(const struct sock *sk) 1308 { 1309 return __sk_stream_memory_free(sk, 0); 1310 } 1311 1312 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1313 { 1314 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1315 __sk_stream_memory_free(sk, wake); 1316 } 1317 1318 static inline bool sk_stream_is_writeable(const struct sock *sk) 1319 { 1320 return __sk_stream_is_writeable(sk, 0); 1321 } 1322 1323 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1324 struct cgroup *ancestor) 1325 { 1326 #ifdef CONFIG_SOCK_CGROUP_DATA 1327 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1328 ancestor); 1329 #else 1330 return -ENOTSUPP; 1331 #endif 1332 } 1333 1334 static inline bool sk_has_memory_pressure(const struct sock *sk) 1335 { 1336 return sk->sk_prot->memory_pressure != NULL; 1337 } 1338 1339 static inline bool sk_under_memory_pressure(const struct sock *sk) 1340 { 1341 if (!sk->sk_prot->memory_pressure) 1342 return false; 1343 1344 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1345 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1346 return true; 1347 1348 return !!*sk->sk_prot->memory_pressure; 1349 } 1350 1351 static inline long 1352 sk_memory_allocated(const struct sock *sk) 1353 { 1354 return atomic_long_read(sk->sk_prot->memory_allocated); 1355 } 1356 1357 static inline long 1358 sk_memory_allocated_add(struct sock *sk, int amt) 1359 { 1360 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1361 } 1362 1363 static inline void 1364 sk_memory_allocated_sub(struct sock *sk, int amt) 1365 { 1366 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1367 } 1368 1369 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1370 1371 static inline void sk_sockets_allocated_dec(struct sock *sk) 1372 { 1373 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1374 SK_ALLOC_PERCPU_COUNTER_BATCH); 1375 } 1376 1377 static inline void sk_sockets_allocated_inc(struct sock *sk) 1378 { 1379 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1380 SK_ALLOC_PERCPU_COUNTER_BATCH); 1381 } 1382 1383 static inline u64 1384 sk_sockets_allocated_read_positive(struct sock *sk) 1385 { 1386 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1387 } 1388 1389 static inline int 1390 proto_sockets_allocated_sum_positive(struct proto *prot) 1391 { 1392 return percpu_counter_sum_positive(prot->sockets_allocated); 1393 } 1394 1395 static inline long 1396 proto_memory_allocated(struct proto *prot) 1397 { 1398 return atomic_long_read(prot->memory_allocated); 1399 } 1400 1401 static inline bool 1402 proto_memory_pressure(struct proto *prot) 1403 { 1404 if (!prot->memory_pressure) 1405 return false; 1406 return !!*prot->memory_pressure; 1407 } 1408 1409 1410 #ifdef CONFIG_PROC_FS 1411 /* Called with local bh disabled */ 1412 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1413 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1414 int sock_inuse_get(struct net *net); 1415 #else 1416 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1417 int inc) 1418 { 1419 } 1420 #endif 1421 1422 1423 /* With per-bucket locks this operation is not-atomic, so that 1424 * this version is not worse. 1425 */ 1426 static inline int __sk_prot_rehash(struct sock *sk) 1427 { 1428 sk->sk_prot->unhash(sk); 1429 return sk->sk_prot->hash(sk); 1430 } 1431 1432 /* About 10 seconds */ 1433 #define SOCK_DESTROY_TIME (10*HZ) 1434 1435 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1436 #define PROT_SOCK 1024 1437 1438 #define SHUTDOWN_MASK 3 1439 #define RCV_SHUTDOWN 1 1440 #define SEND_SHUTDOWN 2 1441 1442 #define SOCK_BINDADDR_LOCK 4 1443 #define SOCK_BINDPORT_LOCK 8 1444 1445 struct socket_alloc { 1446 struct socket socket; 1447 struct inode vfs_inode; 1448 }; 1449 1450 static inline struct socket *SOCKET_I(struct inode *inode) 1451 { 1452 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1453 } 1454 1455 static inline struct inode *SOCK_INODE(struct socket *socket) 1456 { 1457 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1458 } 1459 1460 /* 1461 * Functions for memory accounting 1462 */ 1463 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1464 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1465 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1466 void __sk_mem_reclaim(struct sock *sk, int amount); 1467 1468 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1469 * do not necessarily have 16x time more memory than 4KB ones. 1470 */ 1471 #define SK_MEM_QUANTUM 4096 1472 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1473 #define SK_MEM_SEND 0 1474 #define SK_MEM_RECV 1 1475 1476 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1477 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1478 { 1479 long val = sk->sk_prot->sysctl_mem[index]; 1480 1481 #if PAGE_SIZE > SK_MEM_QUANTUM 1482 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1483 #elif PAGE_SIZE < SK_MEM_QUANTUM 1484 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1485 #endif 1486 return val; 1487 } 1488 1489 static inline int sk_mem_pages(int amt) 1490 { 1491 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1492 } 1493 1494 static inline bool sk_has_account(struct sock *sk) 1495 { 1496 /* return true if protocol supports memory accounting */ 1497 return !!sk->sk_prot->memory_allocated; 1498 } 1499 1500 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1501 { 1502 if (!sk_has_account(sk)) 1503 return true; 1504 return size <= sk->sk_forward_alloc || 1505 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1506 } 1507 1508 static inline bool 1509 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1510 { 1511 if (!sk_has_account(sk)) 1512 return true; 1513 return size <= sk->sk_forward_alloc || 1514 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1515 skb_pfmemalloc(skb); 1516 } 1517 1518 static inline void sk_mem_reclaim(struct sock *sk) 1519 { 1520 if (!sk_has_account(sk)) 1521 return; 1522 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1523 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1524 } 1525 1526 static inline void sk_mem_reclaim_partial(struct sock *sk) 1527 { 1528 if (!sk_has_account(sk)) 1529 return; 1530 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1531 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1532 } 1533 1534 static inline void sk_mem_charge(struct sock *sk, int size) 1535 { 1536 if (!sk_has_account(sk)) 1537 return; 1538 sk->sk_forward_alloc -= size; 1539 } 1540 1541 static inline void sk_mem_uncharge(struct sock *sk, int size) 1542 { 1543 if (!sk_has_account(sk)) 1544 return; 1545 sk->sk_forward_alloc += size; 1546 1547 /* Avoid a possible overflow. 1548 * TCP send queues can make this happen, if sk_mem_reclaim() 1549 * is not called and more than 2 GBytes are released at once. 1550 * 1551 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1552 * no need to hold that much forward allocation anyway. 1553 */ 1554 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1555 __sk_mem_reclaim(sk, 1 << 20); 1556 } 1557 1558 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key); 1559 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1560 { 1561 sk_wmem_queued_add(sk, -skb->truesize); 1562 sk_mem_uncharge(sk, skb->truesize); 1563 if (static_branch_unlikely(&tcp_tx_skb_cache_key) && 1564 !sk->sk_tx_skb_cache && !skb_cloned(skb)) { 1565 skb_ext_reset(skb); 1566 skb_zcopy_clear(skb, true); 1567 sk->sk_tx_skb_cache = skb; 1568 return; 1569 } 1570 __kfree_skb(skb); 1571 } 1572 1573 static inline void sock_release_ownership(struct sock *sk) 1574 { 1575 if (sk->sk_lock.owned) { 1576 sk->sk_lock.owned = 0; 1577 1578 /* The sk_lock has mutex_unlock() semantics: */ 1579 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1580 } 1581 } 1582 1583 /* 1584 * Macro so as to not evaluate some arguments when 1585 * lockdep is not enabled. 1586 * 1587 * Mark both the sk_lock and the sk_lock.slock as a 1588 * per-address-family lock class. 1589 */ 1590 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1591 do { \ 1592 sk->sk_lock.owned = 0; \ 1593 init_waitqueue_head(&sk->sk_lock.wq); \ 1594 spin_lock_init(&(sk)->sk_lock.slock); \ 1595 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1596 sizeof((sk)->sk_lock)); \ 1597 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1598 (skey), (sname)); \ 1599 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1600 } while (0) 1601 1602 static inline bool lockdep_sock_is_held(const struct sock *sk) 1603 { 1604 return lockdep_is_held(&sk->sk_lock) || 1605 lockdep_is_held(&sk->sk_lock.slock); 1606 } 1607 1608 void lock_sock_nested(struct sock *sk, int subclass); 1609 1610 static inline void lock_sock(struct sock *sk) 1611 { 1612 lock_sock_nested(sk, 0); 1613 } 1614 1615 void __lock_sock(struct sock *sk); 1616 void __release_sock(struct sock *sk); 1617 void release_sock(struct sock *sk); 1618 1619 /* BH context may only use the following locking interface. */ 1620 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1621 #define bh_lock_sock_nested(__sk) \ 1622 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1623 SINGLE_DEPTH_NESTING) 1624 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1625 1626 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1627 1628 /** 1629 * unlock_sock_fast - complement of lock_sock_fast 1630 * @sk: socket 1631 * @slow: slow mode 1632 * 1633 * fast unlock socket for user context. 1634 * If slow mode is on, we call regular release_sock() 1635 */ 1636 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1637 __releases(&sk->sk_lock.slock) 1638 { 1639 if (slow) { 1640 release_sock(sk); 1641 __release(&sk->sk_lock.slock); 1642 } else { 1643 spin_unlock_bh(&sk->sk_lock.slock); 1644 } 1645 } 1646 1647 /* Used by processes to "lock" a socket state, so that 1648 * interrupts and bottom half handlers won't change it 1649 * from under us. It essentially blocks any incoming 1650 * packets, so that we won't get any new data or any 1651 * packets that change the state of the socket. 1652 * 1653 * While locked, BH processing will add new packets to 1654 * the backlog queue. This queue is processed by the 1655 * owner of the socket lock right before it is released. 1656 * 1657 * Since ~2.3.5 it is also exclusive sleep lock serializing 1658 * accesses from user process context. 1659 */ 1660 1661 static inline void sock_owned_by_me(const struct sock *sk) 1662 { 1663 #ifdef CONFIG_LOCKDEP 1664 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1665 #endif 1666 } 1667 1668 static inline bool sock_owned_by_user(const struct sock *sk) 1669 { 1670 sock_owned_by_me(sk); 1671 return sk->sk_lock.owned; 1672 } 1673 1674 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1675 { 1676 return sk->sk_lock.owned; 1677 } 1678 1679 /* no reclassification while locks are held */ 1680 static inline bool sock_allow_reclassification(const struct sock *csk) 1681 { 1682 struct sock *sk = (struct sock *)csk; 1683 1684 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1685 } 1686 1687 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1688 struct proto *prot, int kern); 1689 void sk_free(struct sock *sk); 1690 void sk_destruct(struct sock *sk); 1691 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1692 void sk_free_unlock_clone(struct sock *sk); 1693 1694 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1695 gfp_t priority); 1696 void __sock_wfree(struct sk_buff *skb); 1697 void sock_wfree(struct sk_buff *skb); 1698 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1699 gfp_t priority); 1700 void skb_orphan_partial(struct sk_buff *skb); 1701 void sock_rfree(struct sk_buff *skb); 1702 void sock_efree(struct sk_buff *skb); 1703 #ifdef CONFIG_INET 1704 void sock_edemux(struct sk_buff *skb); 1705 void sock_pfree(struct sk_buff *skb); 1706 #else 1707 #define sock_edemux sock_efree 1708 #endif 1709 1710 int sock_setsockopt(struct socket *sock, int level, int op, 1711 sockptr_t optval, unsigned int optlen); 1712 1713 int sock_getsockopt(struct socket *sock, int level, int op, 1714 char __user *optval, int __user *optlen); 1715 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1716 bool timeval, bool time32); 1717 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1718 int noblock, int *errcode); 1719 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1720 unsigned long data_len, int noblock, 1721 int *errcode, int max_page_order); 1722 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1723 void sock_kfree_s(struct sock *sk, void *mem, int size); 1724 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1725 void sk_send_sigurg(struct sock *sk); 1726 1727 struct sockcm_cookie { 1728 u64 transmit_time; 1729 u32 mark; 1730 u16 tsflags; 1731 }; 1732 1733 static inline void sockcm_init(struct sockcm_cookie *sockc, 1734 const struct sock *sk) 1735 { 1736 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1737 } 1738 1739 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1740 struct sockcm_cookie *sockc); 1741 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1742 struct sockcm_cookie *sockc); 1743 1744 /* 1745 * Functions to fill in entries in struct proto_ops when a protocol 1746 * does not implement a particular function. 1747 */ 1748 int sock_no_bind(struct socket *, struct sockaddr *, int); 1749 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1750 int sock_no_socketpair(struct socket *, struct socket *); 1751 int sock_no_accept(struct socket *, struct socket *, int, bool); 1752 int sock_no_getname(struct socket *, struct sockaddr *, int); 1753 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1754 int sock_no_listen(struct socket *, int); 1755 int sock_no_shutdown(struct socket *, int); 1756 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1757 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1758 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1759 int sock_no_mmap(struct file *file, struct socket *sock, 1760 struct vm_area_struct *vma); 1761 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1762 size_t size, int flags); 1763 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1764 int offset, size_t size, int flags); 1765 1766 /* 1767 * Functions to fill in entries in struct proto_ops when a protocol 1768 * uses the inet style. 1769 */ 1770 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1771 char __user *optval, int __user *optlen); 1772 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1773 int flags); 1774 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1775 sockptr_t optval, unsigned int optlen); 1776 1777 void sk_common_release(struct sock *sk); 1778 1779 /* 1780 * Default socket callbacks and setup code 1781 */ 1782 1783 /* Initialise core socket variables */ 1784 void sock_init_data(struct socket *sock, struct sock *sk); 1785 1786 /* 1787 * Socket reference counting postulates. 1788 * 1789 * * Each user of socket SHOULD hold a reference count. 1790 * * Each access point to socket (an hash table bucket, reference from a list, 1791 * running timer, skb in flight MUST hold a reference count. 1792 * * When reference count hits 0, it means it will never increase back. 1793 * * When reference count hits 0, it means that no references from 1794 * outside exist to this socket and current process on current CPU 1795 * is last user and may/should destroy this socket. 1796 * * sk_free is called from any context: process, BH, IRQ. When 1797 * it is called, socket has no references from outside -> sk_free 1798 * may release descendant resources allocated by the socket, but 1799 * to the time when it is called, socket is NOT referenced by any 1800 * hash tables, lists etc. 1801 * * Packets, delivered from outside (from network or from another process) 1802 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1803 * when they sit in queue. Otherwise, packets will leak to hole, when 1804 * socket is looked up by one cpu and unhasing is made by another CPU. 1805 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1806 * (leak to backlog). Packet socket does all the processing inside 1807 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1808 * use separate SMP lock, so that they are prone too. 1809 */ 1810 1811 /* Ungrab socket and destroy it, if it was the last reference. */ 1812 static inline void sock_put(struct sock *sk) 1813 { 1814 if (refcount_dec_and_test(&sk->sk_refcnt)) 1815 sk_free(sk); 1816 } 1817 /* Generic version of sock_put(), dealing with all sockets 1818 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1819 */ 1820 void sock_gen_put(struct sock *sk); 1821 1822 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1823 unsigned int trim_cap, bool refcounted); 1824 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1825 const int nested) 1826 { 1827 return __sk_receive_skb(sk, skb, nested, 1, true); 1828 } 1829 1830 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1831 { 1832 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1833 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1834 return; 1835 sk->sk_tx_queue_mapping = tx_queue; 1836 } 1837 1838 #define NO_QUEUE_MAPPING USHRT_MAX 1839 1840 static inline void sk_tx_queue_clear(struct sock *sk) 1841 { 1842 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1843 } 1844 1845 static inline int sk_tx_queue_get(const struct sock *sk) 1846 { 1847 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1848 return sk->sk_tx_queue_mapping; 1849 1850 return -1; 1851 } 1852 1853 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1854 { 1855 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1856 if (skb_rx_queue_recorded(skb)) { 1857 u16 rx_queue = skb_get_rx_queue(skb); 1858 1859 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1860 return; 1861 1862 sk->sk_rx_queue_mapping = rx_queue; 1863 } 1864 #endif 1865 } 1866 1867 static inline void sk_rx_queue_clear(struct sock *sk) 1868 { 1869 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1870 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1871 #endif 1872 } 1873 1874 static inline int sk_rx_queue_get(const struct sock *sk) 1875 { 1876 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1877 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1878 return sk->sk_rx_queue_mapping; 1879 #endif 1880 1881 return -1; 1882 } 1883 1884 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1885 { 1886 sk->sk_socket = sock; 1887 } 1888 1889 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1890 { 1891 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1892 return &rcu_dereference_raw(sk->sk_wq)->wait; 1893 } 1894 /* Detach socket from process context. 1895 * Announce socket dead, detach it from wait queue and inode. 1896 * Note that parent inode held reference count on this struct sock, 1897 * we do not release it in this function, because protocol 1898 * probably wants some additional cleanups or even continuing 1899 * to work with this socket (TCP). 1900 */ 1901 static inline void sock_orphan(struct sock *sk) 1902 { 1903 write_lock_bh(&sk->sk_callback_lock); 1904 sock_set_flag(sk, SOCK_DEAD); 1905 sk_set_socket(sk, NULL); 1906 sk->sk_wq = NULL; 1907 write_unlock_bh(&sk->sk_callback_lock); 1908 } 1909 1910 static inline void sock_graft(struct sock *sk, struct socket *parent) 1911 { 1912 WARN_ON(parent->sk); 1913 write_lock_bh(&sk->sk_callback_lock); 1914 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1915 parent->sk = sk; 1916 sk_set_socket(sk, parent); 1917 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1918 security_sock_graft(sk, parent); 1919 write_unlock_bh(&sk->sk_callback_lock); 1920 } 1921 1922 kuid_t sock_i_uid(struct sock *sk); 1923 unsigned long sock_i_ino(struct sock *sk); 1924 1925 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1926 { 1927 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1928 } 1929 1930 static inline u32 net_tx_rndhash(void) 1931 { 1932 u32 v = prandom_u32(); 1933 1934 return v ?: 1; 1935 } 1936 1937 static inline void sk_set_txhash(struct sock *sk) 1938 { 1939 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 1940 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 1941 } 1942 1943 static inline bool sk_rethink_txhash(struct sock *sk) 1944 { 1945 if (sk->sk_txhash) { 1946 sk_set_txhash(sk); 1947 return true; 1948 } 1949 return false; 1950 } 1951 1952 static inline struct dst_entry * 1953 __sk_dst_get(struct sock *sk) 1954 { 1955 return rcu_dereference_check(sk->sk_dst_cache, 1956 lockdep_sock_is_held(sk)); 1957 } 1958 1959 static inline struct dst_entry * 1960 sk_dst_get(struct sock *sk) 1961 { 1962 struct dst_entry *dst; 1963 1964 rcu_read_lock(); 1965 dst = rcu_dereference(sk->sk_dst_cache); 1966 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1967 dst = NULL; 1968 rcu_read_unlock(); 1969 return dst; 1970 } 1971 1972 static inline void __dst_negative_advice(struct sock *sk) 1973 { 1974 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1975 1976 if (dst && dst->ops->negative_advice) { 1977 ndst = dst->ops->negative_advice(dst); 1978 1979 if (ndst != dst) { 1980 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1981 sk_tx_queue_clear(sk); 1982 sk->sk_dst_pending_confirm = 0; 1983 } 1984 } 1985 } 1986 1987 static inline void dst_negative_advice(struct sock *sk) 1988 { 1989 sk_rethink_txhash(sk); 1990 __dst_negative_advice(sk); 1991 } 1992 1993 static inline void 1994 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1995 { 1996 struct dst_entry *old_dst; 1997 1998 sk_tx_queue_clear(sk); 1999 sk->sk_dst_pending_confirm = 0; 2000 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2001 lockdep_sock_is_held(sk)); 2002 rcu_assign_pointer(sk->sk_dst_cache, dst); 2003 dst_release(old_dst); 2004 } 2005 2006 static inline void 2007 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2008 { 2009 struct dst_entry *old_dst; 2010 2011 sk_tx_queue_clear(sk); 2012 sk->sk_dst_pending_confirm = 0; 2013 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2014 dst_release(old_dst); 2015 } 2016 2017 static inline void 2018 __sk_dst_reset(struct sock *sk) 2019 { 2020 __sk_dst_set(sk, NULL); 2021 } 2022 2023 static inline void 2024 sk_dst_reset(struct sock *sk) 2025 { 2026 sk_dst_set(sk, NULL); 2027 } 2028 2029 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2030 2031 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2032 2033 static inline void sk_dst_confirm(struct sock *sk) 2034 { 2035 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2036 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2037 } 2038 2039 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2040 { 2041 if (skb_get_dst_pending_confirm(skb)) { 2042 struct sock *sk = skb->sk; 2043 unsigned long now = jiffies; 2044 2045 /* avoid dirtying neighbour */ 2046 if (READ_ONCE(n->confirmed) != now) 2047 WRITE_ONCE(n->confirmed, now); 2048 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2049 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2050 } 2051 } 2052 2053 bool sk_mc_loop(struct sock *sk); 2054 2055 static inline bool sk_can_gso(const struct sock *sk) 2056 { 2057 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2058 } 2059 2060 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2061 2062 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 2063 { 2064 sk->sk_route_nocaps |= flags; 2065 sk->sk_route_caps &= ~flags; 2066 } 2067 2068 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2069 struct iov_iter *from, char *to, 2070 int copy, int offset) 2071 { 2072 if (skb->ip_summed == CHECKSUM_NONE) { 2073 __wsum csum = 0; 2074 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2075 return -EFAULT; 2076 skb->csum = csum_block_add(skb->csum, csum, offset); 2077 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2078 if (!copy_from_iter_full_nocache(to, copy, from)) 2079 return -EFAULT; 2080 } else if (!copy_from_iter_full(to, copy, from)) 2081 return -EFAULT; 2082 2083 return 0; 2084 } 2085 2086 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2087 struct iov_iter *from, int copy) 2088 { 2089 int err, offset = skb->len; 2090 2091 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2092 copy, offset); 2093 if (err) 2094 __skb_trim(skb, offset); 2095 2096 return err; 2097 } 2098 2099 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2100 struct sk_buff *skb, 2101 struct page *page, 2102 int off, int copy) 2103 { 2104 int err; 2105 2106 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2107 copy, skb->len); 2108 if (err) 2109 return err; 2110 2111 skb->len += copy; 2112 skb->data_len += copy; 2113 skb->truesize += copy; 2114 sk_wmem_queued_add(sk, copy); 2115 sk_mem_charge(sk, copy); 2116 return 0; 2117 } 2118 2119 /** 2120 * sk_wmem_alloc_get - returns write allocations 2121 * @sk: socket 2122 * 2123 * Return: sk_wmem_alloc minus initial offset of one 2124 */ 2125 static inline int sk_wmem_alloc_get(const struct sock *sk) 2126 { 2127 return refcount_read(&sk->sk_wmem_alloc) - 1; 2128 } 2129 2130 /** 2131 * sk_rmem_alloc_get - returns read allocations 2132 * @sk: socket 2133 * 2134 * Return: sk_rmem_alloc 2135 */ 2136 static inline int sk_rmem_alloc_get(const struct sock *sk) 2137 { 2138 return atomic_read(&sk->sk_rmem_alloc); 2139 } 2140 2141 /** 2142 * sk_has_allocations - check if allocations are outstanding 2143 * @sk: socket 2144 * 2145 * Return: true if socket has write or read allocations 2146 */ 2147 static inline bool sk_has_allocations(const struct sock *sk) 2148 { 2149 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2150 } 2151 2152 /** 2153 * skwq_has_sleeper - check if there are any waiting processes 2154 * @wq: struct socket_wq 2155 * 2156 * Return: true if socket_wq has waiting processes 2157 * 2158 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2159 * barrier call. They were added due to the race found within the tcp code. 2160 * 2161 * Consider following tcp code paths:: 2162 * 2163 * CPU1 CPU2 2164 * sys_select receive packet 2165 * ... ... 2166 * __add_wait_queue update tp->rcv_nxt 2167 * ... ... 2168 * tp->rcv_nxt check sock_def_readable 2169 * ... { 2170 * schedule rcu_read_lock(); 2171 * wq = rcu_dereference(sk->sk_wq); 2172 * if (wq && waitqueue_active(&wq->wait)) 2173 * wake_up_interruptible(&wq->wait) 2174 * ... 2175 * } 2176 * 2177 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2178 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2179 * could then endup calling schedule and sleep forever if there are no more 2180 * data on the socket. 2181 * 2182 */ 2183 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2184 { 2185 return wq && wq_has_sleeper(&wq->wait); 2186 } 2187 2188 /** 2189 * sock_poll_wait - place memory barrier behind the poll_wait call. 2190 * @filp: file 2191 * @sock: socket to wait on 2192 * @p: poll_table 2193 * 2194 * See the comments in the wq_has_sleeper function. 2195 */ 2196 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2197 poll_table *p) 2198 { 2199 if (!poll_does_not_wait(p)) { 2200 poll_wait(filp, &sock->wq.wait, p); 2201 /* We need to be sure we are in sync with the 2202 * socket flags modification. 2203 * 2204 * This memory barrier is paired in the wq_has_sleeper. 2205 */ 2206 smp_mb(); 2207 } 2208 } 2209 2210 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2211 { 2212 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2213 u32 txhash = READ_ONCE(sk->sk_txhash); 2214 2215 if (txhash) { 2216 skb->l4_hash = 1; 2217 skb->hash = txhash; 2218 } 2219 } 2220 2221 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2222 2223 /* 2224 * Queue a received datagram if it will fit. Stream and sequenced 2225 * protocols can't normally use this as they need to fit buffers in 2226 * and play with them. 2227 * 2228 * Inlined as it's very short and called for pretty much every 2229 * packet ever received. 2230 */ 2231 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2232 { 2233 skb_orphan(skb); 2234 skb->sk = sk; 2235 skb->destructor = sock_rfree; 2236 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2237 sk_mem_charge(sk, skb->truesize); 2238 } 2239 2240 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2241 { 2242 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2243 skb_orphan(skb); 2244 skb->destructor = sock_efree; 2245 skb->sk = sk; 2246 return true; 2247 } 2248 return false; 2249 } 2250 2251 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2252 { 2253 if (skb->destructor != sock_wfree) { 2254 skb_orphan(skb); 2255 return; 2256 } 2257 skb->slow_gro = 1; 2258 } 2259 2260 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2261 unsigned long expires); 2262 2263 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2264 2265 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2266 2267 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2268 struct sk_buff *skb, unsigned int flags, 2269 void (*destructor)(struct sock *sk, 2270 struct sk_buff *skb)); 2271 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2272 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2273 2274 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2275 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2276 2277 /* 2278 * Recover an error report and clear atomically 2279 */ 2280 2281 static inline int sock_error(struct sock *sk) 2282 { 2283 int err; 2284 2285 /* Avoid an atomic operation for the common case. 2286 * This is racy since another cpu/thread can change sk_err under us. 2287 */ 2288 if (likely(data_race(!sk->sk_err))) 2289 return 0; 2290 2291 err = xchg(&sk->sk_err, 0); 2292 return -err; 2293 } 2294 2295 void sk_error_report(struct sock *sk); 2296 2297 static inline unsigned long sock_wspace(struct sock *sk) 2298 { 2299 int amt = 0; 2300 2301 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2302 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2303 if (amt < 0) 2304 amt = 0; 2305 } 2306 return amt; 2307 } 2308 2309 /* Note: 2310 * We use sk->sk_wq_raw, from contexts knowing this 2311 * pointer is not NULL and cannot disappear/change. 2312 */ 2313 static inline void sk_set_bit(int nr, struct sock *sk) 2314 { 2315 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2316 !sock_flag(sk, SOCK_FASYNC)) 2317 return; 2318 2319 set_bit(nr, &sk->sk_wq_raw->flags); 2320 } 2321 2322 static inline void sk_clear_bit(int nr, struct sock *sk) 2323 { 2324 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2325 !sock_flag(sk, SOCK_FASYNC)) 2326 return; 2327 2328 clear_bit(nr, &sk->sk_wq_raw->flags); 2329 } 2330 2331 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2332 { 2333 if (sock_flag(sk, SOCK_FASYNC)) { 2334 rcu_read_lock(); 2335 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2336 rcu_read_unlock(); 2337 } 2338 } 2339 2340 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2341 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2342 * Note: for send buffers, TCP works better if we can build two skbs at 2343 * minimum. 2344 */ 2345 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2346 2347 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2348 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2349 2350 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2351 { 2352 u32 val; 2353 2354 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2355 return; 2356 2357 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2358 2359 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2360 } 2361 2362 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2363 bool force_schedule); 2364 2365 /** 2366 * sk_page_frag - return an appropriate page_frag 2367 * @sk: socket 2368 * 2369 * Use the per task page_frag instead of the per socket one for 2370 * optimization when we know that we're in the normal context and owns 2371 * everything that's associated with %current. 2372 * 2373 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2374 * inside other socket operations and end up recursing into sk_page_frag() 2375 * while it's already in use. 2376 * 2377 * Return: a per task page_frag if context allows that, 2378 * otherwise a per socket one. 2379 */ 2380 static inline struct page_frag *sk_page_frag(struct sock *sk) 2381 { 2382 if (gfpflags_normal_context(sk->sk_allocation)) 2383 return ¤t->task_frag; 2384 2385 return &sk->sk_frag; 2386 } 2387 2388 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2389 2390 /* 2391 * Default write policy as shown to user space via poll/select/SIGIO 2392 */ 2393 static inline bool sock_writeable(const struct sock *sk) 2394 { 2395 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2396 } 2397 2398 static inline gfp_t gfp_any(void) 2399 { 2400 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2401 } 2402 2403 static inline gfp_t gfp_memcg_charge(void) 2404 { 2405 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2406 } 2407 2408 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2409 { 2410 return noblock ? 0 : sk->sk_rcvtimeo; 2411 } 2412 2413 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2414 { 2415 return noblock ? 0 : sk->sk_sndtimeo; 2416 } 2417 2418 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2419 { 2420 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2421 2422 return v ?: 1; 2423 } 2424 2425 /* Alas, with timeout socket operations are not restartable. 2426 * Compare this to poll(). 2427 */ 2428 static inline int sock_intr_errno(long timeo) 2429 { 2430 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2431 } 2432 2433 struct sock_skb_cb { 2434 u32 dropcount; 2435 }; 2436 2437 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2438 * using skb->cb[] would keep using it directly and utilize its 2439 * alignement guarantee. 2440 */ 2441 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2442 sizeof(struct sock_skb_cb))) 2443 2444 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2445 SOCK_SKB_CB_OFFSET)) 2446 2447 #define sock_skb_cb_check_size(size) \ 2448 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2449 2450 static inline void 2451 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2452 { 2453 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2454 atomic_read(&sk->sk_drops) : 0; 2455 } 2456 2457 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2458 { 2459 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2460 2461 atomic_add(segs, &sk->sk_drops); 2462 } 2463 2464 static inline ktime_t sock_read_timestamp(struct sock *sk) 2465 { 2466 #if BITS_PER_LONG==32 2467 unsigned int seq; 2468 ktime_t kt; 2469 2470 do { 2471 seq = read_seqbegin(&sk->sk_stamp_seq); 2472 kt = sk->sk_stamp; 2473 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2474 2475 return kt; 2476 #else 2477 return READ_ONCE(sk->sk_stamp); 2478 #endif 2479 } 2480 2481 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2482 { 2483 #if BITS_PER_LONG==32 2484 write_seqlock(&sk->sk_stamp_seq); 2485 sk->sk_stamp = kt; 2486 write_sequnlock(&sk->sk_stamp_seq); 2487 #else 2488 WRITE_ONCE(sk->sk_stamp, kt); 2489 #endif 2490 } 2491 2492 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2493 struct sk_buff *skb); 2494 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2495 struct sk_buff *skb); 2496 2497 static inline void 2498 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2499 { 2500 ktime_t kt = skb->tstamp; 2501 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2502 2503 /* 2504 * generate control messages if 2505 * - receive time stamping in software requested 2506 * - software time stamp available and wanted 2507 * - hardware time stamps available and wanted 2508 */ 2509 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2510 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2511 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2512 (hwtstamps->hwtstamp && 2513 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2514 __sock_recv_timestamp(msg, sk, skb); 2515 else 2516 sock_write_timestamp(sk, kt); 2517 2518 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2519 __sock_recv_wifi_status(msg, sk, skb); 2520 } 2521 2522 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2523 struct sk_buff *skb); 2524 2525 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2526 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2527 struct sk_buff *skb) 2528 { 2529 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2530 (1UL << SOCK_RCVTSTAMP)) 2531 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2532 SOF_TIMESTAMPING_RAW_HARDWARE) 2533 2534 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2535 __sock_recv_ts_and_drops(msg, sk, skb); 2536 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2537 sock_write_timestamp(sk, skb->tstamp); 2538 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2539 sock_write_timestamp(sk, 0); 2540 } 2541 2542 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2543 2544 /** 2545 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2546 * @sk: socket sending this packet 2547 * @tsflags: timestamping flags to use 2548 * @tx_flags: completed with instructions for time stamping 2549 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2550 * 2551 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2552 */ 2553 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2554 __u8 *tx_flags, __u32 *tskey) 2555 { 2556 if (unlikely(tsflags)) { 2557 __sock_tx_timestamp(tsflags, tx_flags); 2558 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2559 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2560 *tskey = sk->sk_tskey++; 2561 } 2562 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2563 *tx_flags |= SKBTX_WIFI_STATUS; 2564 } 2565 2566 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2567 __u8 *tx_flags) 2568 { 2569 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2570 } 2571 2572 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2573 { 2574 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2575 &skb_shinfo(skb)->tskey); 2576 } 2577 2578 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key); 2579 /** 2580 * sk_eat_skb - Release a skb if it is no longer needed 2581 * @sk: socket to eat this skb from 2582 * @skb: socket buffer to eat 2583 * 2584 * This routine must be called with interrupts disabled or with the socket 2585 * locked so that the sk_buff queue operation is ok. 2586 */ 2587 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2588 { 2589 __skb_unlink(skb, &sk->sk_receive_queue); 2590 if (static_branch_unlikely(&tcp_rx_skb_cache_key) && 2591 !sk->sk_rx_skb_cache) { 2592 sk->sk_rx_skb_cache = skb; 2593 skb_orphan(skb); 2594 return; 2595 } 2596 __kfree_skb(skb); 2597 } 2598 2599 static inline 2600 struct net *sock_net(const struct sock *sk) 2601 { 2602 return read_pnet(&sk->sk_net); 2603 } 2604 2605 static inline 2606 void sock_net_set(struct sock *sk, struct net *net) 2607 { 2608 write_pnet(&sk->sk_net, net); 2609 } 2610 2611 static inline bool 2612 skb_sk_is_prefetched(struct sk_buff *skb) 2613 { 2614 #ifdef CONFIG_INET 2615 return skb->destructor == sock_pfree; 2616 #else 2617 return false; 2618 #endif /* CONFIG_INET */ 2619 } 2620 2621 /* This helper checks if a socket is a full socket, 2622 * ie _not_ a timewait or request socket. 2623 */ 2624 static inline bool sk_fullsock(const struct sock *sk) 2625 { 2626 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2627 } 2628 2629 static inline bool 2630 sk_is_refcounted(struct sock *sk) 2631 { 2632 /* Only full sockets have sk->sk_flags. */ 2633 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2634 } 2635 2636 /** 2637 * skb_steal_sock - steal a socket from an sk_buff 2638 * @skb: sk_buff to steal the socket from 2639 * @refcounted: is set to true if the socket is reference-counted 2640 */ 2641 static inline struct sock * 2642 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2643 { 2644 if (skb->sk) { 2645 struct sock *sk = skb->sk; 2646 2647 *refcounted = true; 2648 if (skb_sk_is_prefetched(skb)) 2649 *refcounted = sk_is_refcounted(sk); 2650 skb->destructor = NULL; 2651 skb->sk = NULL; 2652 return sk; 2653 } 2654 *refcounted = false; 2655 return NULL; 2656 } 2657 2658 /* Checks if this SKB belongs to an HW offloaded socket 2659 * and whether any SW fallbacks are required based on dev. 2660 * Check decrypted mark in case skb_orphan() cleared socket. 2661 */ 2662 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2663 struct net_device *dev) 2664 { 2665 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2666 struct sock *sk = skb->sk; 2667 2668 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2669 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2670 #ifdef CONFIG_TLS_DEVICE 2671 } else if (unlikely(skb->decrypted)) { 2672 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2673 kfree_skb(skb); 2674 skb = NULL; 2675 #endif 2676 } 2677 #endif 2678 2679 return skb; 2680 } 2681 2682 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2683 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2684 */ 2685 static inline bool sk_listener(const struct sock *sk) 2686 { 2687 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2688 } 2689 2690 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2691 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2692 int type); 2693 2694 bool sk_ns_capable(const struct sock *sk, 2695 struct user_namespace *user_ns, int cap); 2696 bool sk_capable(const struct sock *sk, int cap); 2697 bool sk_net_capable(const struct sock *sk, int cap); 2698 2699 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2700 2701 /* Take into consideration the size of the struct sk_buff overhead in the 2702 * determination of these values, since that is non-constant across 2703 * platforms. This makes socket queueing behavior and performance 2704 * not depend upon such differences. 2705 */ 2706 #define _SK_MEM_PACKETS 256 2707 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2708 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2709 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2710 2711 extern __u32 sysctl_wmem_max; 2712 extern __u32 sysctl_rmem_max; 2713 2714 extern int sysctl_tstamp_allow_data; 2715 extern int sysctl_optmem_max; 2716 2717 extern __u32 sysctl_wmem_default; 2718 extern __u32 sysctl_rmem_default; 2719 2720 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2721 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2722 2723 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2724 { 2725 /* Does this proto have per netns sysctl_wmem ? */ 2726 if (proto->sysctl_wmem_offset) 2727 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2728 2729 return *proto->sysctl_wmem; 2730 } 2731 2732 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2733 { 2734 /* Does this proto have per netns sysctl_rmem ? */ 2735 if (proto->sysctl_rmem_offset) 2736 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2737 2738 return *proto->sysctl_rmem; 2739 } 2740 2741 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2742 * Some wifi drivers need to tweak it to get more chunks. 2743 * They can use this helper from their ndo_start_xmit() 2744 */ 2745 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2746 { 2747 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2748 return; 2749 WRITE_ONCE(sk->sk_pacing_shift, val); 2750 } 2751 2752 /* if a socket is bound to a device, check that the given device 2753 * index is either the same or that the socket is bound to an L3 2754 * master device and the given device index is also enslaved to 2755 * that L3 master 2756 */ 2757 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2758 { 2759 int mdif; 2760 2761 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2762 return true; 2763 2764 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2765 if (mdif && mdif == sk->sk_bound_dev_if) 2766 return true; 2767 2768 return false; 2769 } 2770 2771 void sock_def_readable(struct sock *sk); 2772 2773 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2774 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2775 int sock_set_timestamping(struct sock *sk, int optname, 2776 struct so_timestamping timestamping); 2777 2778 void sock_enable_timestamps(struct sock *sk); 2779 void sock_no_linger(struct sock *sk); 2780 void sock_set_keepalive(struct sock *sk); 2781 void sock_set_priority(struct sock *sk, u32 priority); 2782 void sock_set_rcvbuf(struct sock *sk, int val); 2783 void sock_set_mark(struct sock *sk, u32 val); 2784 void sock_set_reuseaddr(struct sock *sk); 2785 void sock_set_reuseport(struct sock *sk); 2786 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2787 2788 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2789 2790 #endif /* _SOCK_H */ 2791